Skip to content

Multimodal Fallacy Classification in Political Debates: Dataset and Experiments.

License

Notifications You must be signed in to change notification settings

nlp-unibo/multimodal-am-fallacy

Repository files navigation

Multimodal Fallacy Classification in Political Debates

This repository contains the code and resources for the project "Multimodal Argument Mining: A Case Study in Political Debates," focusing on the exploration of audio's role in classifying argumentative fallacies in political debates.

Dataset: MM-USED-fallacies

The repository includes the code required to generate the MM-USED-fallacies dataset. This dataset is created by leveraging resources from the MM-USED dataset and the USED-fallacy dataset. By incorporating multimodal techniques, we aim to enrich the fallacy analysis beyond traditional text-based information, specifically by incorporating audio data from political debates.

Repository Structure

The main directory includes the following subdirectories:

  • local_database/: This directory contains the original datasets USED-fallacy and MM-DatasetFallacies. It includes three versions of the MM-DatasetFallacies dataset: full, no_duplicates, and partial-used (for debugging purposes).
  • resources/: This directory contains additional resources for the experiments.
    • clips_generation/: This subdirectory contains the files necessary to generate audio clips corresponding to dialogues, snippets, and components.
    • download/: This subdirectory contains the files necessary to download the recordings used in the experiments.
  • results/: This directory stores the results of the experiments.
  • runnables/: This directory houses the scripts necessary to run the experiments.
  • utils/: This directory contains utility files for the experiments.

Please refer to the specific directories for further details on their contents and usage.

Usage

  1. Dataset Generation:

    • To generate the MM-USED-fallacy dataset starting from MM-USED and USED-fallacy, follow the following steps. The code and resources necessary for the dataset generation are included. Please, note that the file containing all the annotations for the MM-USED-fallacy dataset is already included in the local_database/MM-DatasetFallacies/full directory under the name dataset.csv.
      • Download the recordings of the debates using the information provided in the experiments/resources/download/download_links.csv file. The recordings should saved under resources/debates_audio_recordings. Please, use whatever tool you prefer to obtain the recordings.
      • Run the run_clips_generation.sh script in the runnables directory create the audio clips corresponding to the dialogues, snippets, and components. The script will generate the MM-USED-fallacy dataset clips and store them under a new folder local_database/MM-DatasetFallacies/audio_clips.
  2. Experiments:

    • To run the experiments in the paper and perform leave-one-debate-out cross-validation:
      • Open the script runnables/leave_one_out.py and set:
        • text_model: can have values bert, roberta, sbert
        • audio model: can have values wav2vec, clap
        • config: can have values text_only, audio_only, text_audio
      • Run the run_leave_one_out.sh script in the runnables directory. The script will run the experiments for each debate and store the results in the results directory.

Please refer to the specific directories for further details on their contents and usage.

License

This project is licensed under the CC.BY License.

Citing the work

If using this dataset, please cite the following publication:

Eleonora Mancini, Federico Ruggeri, and Paolo Torroni. 2024. Multimodal Fallacy Classification in Political Debates. In Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 2: Short Papers), pages 170–178, St. Julian’s, Malta. Association for Computational Linguistics.

@inproceedings{mancini-etal-2024-multimodal,
    title = "Multimodal Fallacy Classification in Political Debates",
    author = "Mancini, Eleonora  and
      Ruggeri, Federico  and
      Torroni, Paolo",
    editor = "Graham, Yvette  and
      Purver, Matthew",
    booktitle = "Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 2: Short Papers)",
    month = mar,
    year = "2024",
    address = "St. Julian{'}s, Malta",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.eacl-short.16",
    pages = "170--178",
}

About

Multimodal Fallacy Classification in Political Debates: Dataset and Experiments.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published